View
6
Download
0
Category
Preview:
Citation preview
Empirical Evidence from a Japanese Lending Survey
within the TVP-VAR Framework:
Does the Credit Channel Matter for Monetary Policy?
Tatsuki Okamoto
Yoichi Matsubayashi
March 2017
Discussion Paper No.1709
GRADUATE SCHOOL OF ECONOMICS
KOBE UNIVERSITY
ROKKO, KOBE, JAPAN
1
Empirical Evidence from a Japanese Lending Survey within the TVP-VAR Framework:
Does the Credit Channel Matter for Monetary Policy?
Tatsuki Okamoto
Graduate School of Economics, Kobe University
Yoichi Matsubayashi
Graduate School of Economics, Kobe University
Abstract This paper examines whether Japanese monetary policy had been working through the credit
channel and its sub-channels between March 2000 and March 2016 using time-varying
parameter VAR. The identification of credit transmission channels is a very difficult problem
due to the impossibility to observe the conditions of credit supply and demand. However, using
the credible data collected from the ‘Senior Loan Officer Opinion Survey on Bank Lending
Practices at Large Japanese Banks’ (SLOS), we identified the credit channel and its
sub-channels. To the best of the authors’ knowledge, there are no previous studies that have
employed SLOS data for the evaluation of transmission channels. The estimation findings show
a high possibility that large and middle-sized firms had little effect on monetary policy through
the credit channel, but did have an effect through portfolio rebalancing. Small firms are thought
to have an effect through the credit channel and its sub-channels, but it is not a big effect. The
detailed reason as to why the effect of monetary easing differed by the firm size should be
considered by looking at more specific portfolio rebalancing effects and loans to overseas.
Keywords: Time-Varying Parameter vector autoregressive (TVP-VAR) model, Credit Channel,
Credit supply, Lending standards, Monetary policy.
JEL classification: E41; E44; E51; E52; G21
Correspondence to:
Tatsuki Okamoto
Graduate School of Economics, Kobe University
Rokkodai, Nada-ku, Kobe, Hyogo 657-8501, Japan
TEL +81-090-1240-9543
E-mail: tatsuki.okamoto.82@gmail.com
2
1 Introduction
Some preceding studies (Honda et al., 2007; Ijiri, 2016) suggest that Japanese monetary
policy from 2001 to 2006 did not work much through the banking sector, in other words, Credit
Channel1. However, these studies did not consider the determinants of loans. Based on this fact,
this paper tries to disentangle the determinants of loans into loan demand and lending standards
to understand whether those factors really affected the amounts of loans. Here, we break the
credit channel into two sub-channels: the Bank Incentive Channel (BIC) and the Bank Observe
Channel (BOC) following the way of Ciccarelli et al. (2015)2. These sub-channels express bank
factor and corporate factor. As the main contribution, we verify the effect of monetary policy
using Japanese SLOS and the time-varying parameter VAR (TVP-VAR). Then, we use
TVP-VAR for estimation, because we aim to assess the effect of monetary policy in several
policy terms including the zero lower bound (ZLB). In this way, the paper tries to analyse the
extent to which Japan’s monetary policy affected Indices of Industrial Production (IIP) through
the credit channel from 2000 to 2016 depending on the policy terms. In particular, we focus on
the following four questions: (1) Did Japanese monetary policy really affect IIP through the
credit channel? (2) Which factors explain the loan amount fluctuation, loan demand and lending
standard? (3) If lending standards significantly affect loans, then specifically which factor
affected this effect, the bank factor or the corporate factor? (4) Do the effects from these factors
on loans differ depending on the firm size or the policy term? To answer these questions, we try
to identify the credit channel by using data from the ‘Senior Loan Officer Opinion Survey on
Bank Lending Practices at Large Japanese Banks’ (SLOS), which allows us to disentangle the
factors3. The results showed that the effect of monetary policy and the scale of the credit
channel differ depending on the firm size. Table 1 reports the summary of monetary
transmission channels in this paper.
[Table 1]
This paper is structured as follows. Section 2 explains the transmission channel and the
framework of channels. Section 3 describes SLOS data and how to calculate the channel
variables. Section 4 summarises the data used in this paper. Section 5 reviews the models.
Section 6 closely looks at the estimation results. Section 7 presents the conclusions.
1 Bernanke and Blinder (1988, 1992), Kashyap and Stein (1994) and Bernanke and Gertler (1995)
describe Credit Channel and its sub-channels in detail. 2 According to Ciccarelli et al. (2015), which used FRB’s SLOS and ECB’s BLS, it is difficult to identify
the credit channel and its sub-channels with loans due to restricted information about the loan demand
and the supply; however, the approach of this paper works well to identify them. 3 Basically, ‘bank’ in this paper refers to financial institutions in general, including ‘shinkin’ and others.
3
2 Transmission channels In this paper, we follow Ciccarelli et al. (2015) and regard the lending standard as the broad
credit channel (BCC) variable in order to disentangle the sub-channels of BCC into the bank
incentive channel (BIC) and bank observe channel (BOC). This approach is different from
Ciccarelli et al. (2015) as they regard fluctuation of loans as BCC whereas we regard them as
different variables. We explain each variable below.
First, the cost of capital channel (CCC or demand) expresses the variable of loan demand as
described by Ciccarelli et al. (2015). Observing this channel allows us to determine how firms’
real economic activities and financing are affected by the monetary policy.
Second, the broad credit channel (BCC) is also named following Ciccarelli et al. (2015).
BCC is thought to be a variable that allows us to capture the fluctuation of the lending standards.
For example, banks react to the monetary policy and ease their lending standards by easing the
condition of collaterals and real interest rate. In this paper, we regard this variable different from
loans because the amount of loans might be determined by the loan demand and the lending
standard. BCC is more flexible variable than loans at this point. Moreover, we use this variable
to disentangle BCC into two sub-channels: BIC and BOC.
Third, the BIC expresses how much the bank’s balance sheet expansions explain the
fluctuation of lending standard. BIC represents the size of bankers’ emotions. This corresponds
to the bank lending channel (or bank assets channel) of Bernanke and Gertler (1995); however,
we named this variable ‘bank incentive channel’ to differentiate from the bank lending channel,
which often uses the stock price of banks as an alternative variable that includes expectations.
BIC strongly relates with portfolio rebalancing4. This is because it is expected that the Bank of
Japan (BOJ) purchases government bonds from commercial banks, reserve deposit balances
(high-powered money) increase and the compositions of the banks’ balance sheets change. Then,
banks have some choices to deal with changes in the portfolios of their balance sheets: (1)
increase loans; (2) repurchase bonds; or (3) increase stocks. One or some of them will be taken
into practise from these choices. Here, the expansion in BIC means the ratio of loans in banks’
portfolios increase or the asset price in the banks’ balance sheets increase.
Finally, the BOC expresses the extent to which the banks’ prospects of firms explain the
fluctuation of lending standard. The BOC is a psychological variable of the banks. At this point,
this does not fit with the balance sheet variable of Bernanke and Gertler (1995) or Oliner and
Rudebusch (1996).
The framework of monetary policy transmission channel in this paper should be referred in
Figure 1.
[Figure 1]
In the next chapter, we will explain how those variables are derived from the SLOS data.
4 Bernanke and Reinhart (2004) explain portfolio rebalancing effect in detail.
4
3 Survey The unique point of this paper is that estimates have been done with some variables created
by the authors from SLOS data. Following Ciccarelli et al. (2015), these variables are calculated
and used for estimation to analyse the effects of monetary policy at a macro level. In other
words, by using these variables, we can determine whether changes in lending standards are due
to bank assets or due to banks’ views of firms’ business prospects. In this way, we can also
determine the effectiveness of each channel. In an attempt to evaluate these channels, many
empirical analyses have been conducted using variables created from the questionnaire data.
However, most of these studies use the BOJ’s ‘Tankan’, which incorporates the viewpoint of
firms5. Analysing the credit channel with SLOS is thought to be much more accurate than using
Tankan or each firm’s micro data. This is because SLOS is answered by bankers who fully
understand the capital situation of their own financial institutions, which means we can capture
the macro economy from the financial institution’s perspective with these data. As in Figure 2,
which shows the movement of BCC created from SLOS and Lending Attitude calculated from
Tankan, we can observe that the sentiments of demand side and supply side on loans are largely
different after 2014. To the best of the authors’ knowledge, there is no study that has used SLOS
data for the evaluation of transmission channels.
[Figure 2]
The SLOS data are obtained from the BOJ’s website. The survey began in April 2000 and
the answers are taken quarterly by the BOJ from 50 Japanese large private banks. These
respondents are chosen from the viewpoint of lending volume. As mentioned in the BOJ’s
website, ‘the aggregate loan size of these banks account for approximately 75 percent of the
loan market of Japanese private banks (city banks, regional banks, regional banks II, trust banks,
long-term credit banks and shinkin banks)’. In addition, respondents of this survey are
reconsidered every 3 years. The survey includes several questions related to changes in loan
demand and changes in lending standards to several types of firms. Firms are mainly classified
as small, medium or large. Banks are also asked to provide the reason why they changed their
lending standard by rating seven possible reasons on a three-point scale: 3 = important, 2 =
somewhat important, 1 = not important. The BOJ takes a weighted average of these ratings
while considering banks’ share and the scale of products (see the BOJ’s website for detailed
information).
The way of making CCC and BCC follows Lown and Morgan (2006) and Ciccarelli et al.
(2015). They used FRB’s SLOS and ECB’s BLS. Regarding the BIC and BOC, the way of
creating them is considered by the authors. The BIC and BOC are divided by 100 to match the
size with the CCC and BCC. We will explain the way of calculating each channel.
5 For example, Ogawa (2003) uses the Tankan’s ‘banks’ Lending Attitude’.
5
The CCC is created from Q26.
CCC = (percentage of respondents selecting ‘substantially stronger’ + ‘moderately stronger’) –
(percentage of respondents selecting ‘moderately weaker’ + ‘substantially weaker’) (1)
The BCC is created from Q77.
BCC = (percentage of respondents selecting ‘eased considerably’ + ‘eased somewhat’) – (percentage of
respondents selecting ‘tightened somewhat’ + ‘tightened considerably’) (2)
The BIC and BOC are created from Q7, Q8.a. and Q8.b.
Regarding Q8, (ⅰ)–(ⅲ) are as follows: (ⅰ) An improvement (or deterioration) in your bank’s
asset portfolio; (ⅱ) A more (or less) favourable or less (more) uncertain economic outlook; (ⅲ)
An improvement in (or worsening of) industry or firm specific problems8.
BIC = {(𝛼1× Number of banks answering ‘eased’ in Q7) − (𝛽1× Number of banks answering ‘tightened’
in Q7)}/100 (3)
BOC = {(𝛼23×Number of banks answering ‘eased’ in Q7) − (𝛽23×Number of banks answering
‘tightened’ in Q7)}/100 (4)
Here, 𝜶𝟏, 𝜶𝟐𝟑, 𝜷𝟏 and 𝜷𝟐𝟑 are as follows:
𝜶𝟏: scale value weighted averaged of (ⅰ) in Q8.a.
𝜶𝟐𝟑: sum of scale value weighted averaged of (ⅱ) and (ⅲ) in Q8.a. divided by 2.
𝜷𝟏: scale value weighted averaged of (ⅰ) in Q8.b.
𝜷𝟐𝟑: sum of scale value weighted averaged of (ⅱ) and (ⅲ) in Q8.b. divided by 2.
The availability of these variables should be discussed because banks have no incentive to
answer honestly. In the preceding studies, the availability of data made by SLOS or BLS is
checked roughly in two ways. The first involves applying a multi-regression model. The second
is to check the correlation between the loan and BCC. The details of the regression and ways to
confirm the availability should be referred to Lown et al. (2000), Lown and Morgan (2006),
Maddaloni and Peydró (2011), Del Giovane et al. (2011) and Bassett et al. (2014). These studies
suggest that the data from SLOS or BLS have a significant effect on forecasting loans and GDP.
Focussing on Japanese data made from SLOS, Kano (2006) refers to its reliability and
credibility. Outside of this paper, we performed a multi-regression using the method of Newey
6 Q2. How has demand for loans from firms changed over the past 3 months according to the industry
and the firm size? 7 Q7. Over the past 3 months, how have your bank’s credit standards for approving applications for loans
from firms and households changed? 8 Q8.a. If your bank has eased its credit standards for loans to firms (that is, the answer to question 7 is
either ‘eased considerably’ or ‘eased somewhat’), to what factors do you attribute this easing? (Please rate
each possible reason using the following scale: 3 = important, 2 = somewhat important, 1 = not
important.)
Q8.b. If your bank has tightened its credit standards for loans to firms over the past 3 months (as
described in question 7), what were the important factors that led to the change? (Please rate each
possible reason using the following scale: 3 = important, 2 = somewhat important, 1 = not important.)
6
and West (1987) and found that within a 15% significance level, the CCC and BCC have a
significant effect on loans. However, the specific point of this paper is that we do not regard
BCC as an alternative variable of loan. This means that it is not necessary to check the
availability of BCC as an alternative variable of loan. Therefore, based on this evidence, we use
these data in a TVP-VAR regression.
7
4 Data and Models
4-1. Data We use quarterly data between March 2000 and March 2016. This is because the data
sampling of SLOS started in April 2000. SLOS includes the different sampling periods, though
SLOS asks about the last 3 months’ fluctuation; hence, we regard the sample period of SLOS
corresponds to the other data. Another specific point is that we use IIP instead of GDP. This is
because IIP is expected to make the relationship between the monetary policy and loans for
investment much clearer than GDP. The summary of data and variables utilised in the estimation
are described in Table 2. It should be noted that these variables are seasonally adjusted by
E-views8, ARIMA X-12 and all data are demeaned. For Y, L and M, logs are taken before they
are demeaned. In this paper, firm size categorisation follows that of the BOJ (see the BOJ’s
website for details).
[Table 2]
4-2. Models
We estimate five models: Model 1 (L,Y,M), Model 2 (BCC,L,Y,M), Model 3 (CCC,L,Y,M),
Model 4 (BIC,L,Y,M) and Model 5 (BOC,L,Y,M) for each firm size (large, middle and small).
The details of the variables are described in Table 1.
In Model 1 (L,Y,M), we check the effect of monetary policy to loans. Then, in Model 2
(BCC,L,Y,M), we observe the effect of monetary policy on loans considering the channel
through the lending standard. In estimating Model 3 (CCC,L,Y,M), we test the hypothesis that
the monetary policy did not affect the real economy through the channel of loans to IIP because
of its invalid effect on loan demand. Lastly, from Model 4 (BIC,L,Y,M) and Model 5
(BOC,L,Y,M), we consider whether bank problems, such as bad loans and non-performing loans,
or low capital adequacy ratio caused the low growth of loan, or the bad prospects on firms
caused this problem.
There are some cautions in these models: (1) the sample period includes ZLB; (2) we do not
use the interest rate or the monetary bond as a monetary policy variable and (3) the price
variable is not concerned in this paper. The first point recognises the sample period to include
the zero lower bound (ZLB) period. Nakajima (2011b) describes the way to deal with this
problem by utilising the Tobit-type non-linearity method to the nominal interest rate to make it a
censored variable9. However, as a result of using this method in the framework of TVP-VAR,
Nakajima (2011b) concluded that there are not any big difference in the result of models with or
without considering the ZLB10
. Based on this finding, in this paper, we make ZLB as a problem
to tackle in the future and we will estimate without using the Tobit-type non-linearity
method11
.The second problem to tackle is selecting the monetary policy variable dealing with
9 Details are also in Iwata and Wu (2006).
10 Details should be referred to Nakajima (2011b).
11 Other papers such as Nakajima and West (2013) and Kimura and Nakajima (2016) deal with the ZLB
period by utilising the ‘latent threshold model (LTM)’ to let the model change its regime between
8
the fact that the monetary policy differs in every policy term. Regarding TVP-VAR, this method
has a restriction with its number of variables in the model. Thus, we deal with this problem
using the CAB as a monetary policy variable12
. We also do not consider the interest rate in this
model for the following reasons: (1) the short-term interest rate in this sample period
transitioned around 0% and (2) Honda et al. (2007) mention that in reacting to the monetary
policy, the long-term interest rate did not go down, but rather increased during 2001–2006. The
last point is that we do not use a price variable for the estimation for the following reasons. First,
the number of variables that can be used in TVP-VAR is limited. Second, Honda et al. (2007),
which estimates for the quantitative easing period from 2001 to 2006, states that the consumer
price response to the monetary policy shock is very small and not significantly far from zero in
the whole period. This means the effect of monetary easing policy on price is uncertain. Lastly,
Williams (2012) of the Federal Reserve Bank of San Francisco at the time also stated that the
‘reserve deposit balance hardly affects money stock, bank lending and inflation’.
In this sub-chapter, we will explain the orders of variables in detail. The order utilised in this
paper is different from the one in Christiano et al. (1999). This is because we set the following
assumptions in the models. First, we assume that the monetary policy is determined after the
central bank observes IIP (Y), loans (L) and SLOS (C). In other words, it is assumed that
monetary policy influences other variables one term later. Second, we assume private banks
individually decide loan amount (L) while observing loan demand, prospects of firms and
changes in bank assets(C). Lastly, private banks are assumed that they can observe changes in
loan demand and changes in bank assets(C) prior to IIP (Y) and that a private bank can
determine loan (L) before IIP (Y) is observed. Originally, this order should be determined using
the marginal likelihood method. In this paper, however, checking some orders, such as (Y,L,M),
(Y,C,L,M) and (Y,L,C,M), did not reveal any big difference; thus, we adopt the order (C,L,Y,M).
conventional and unconventional policies. 12
Refer to Kimura and Nakajima (2016) about why we can use CAB on this terms’ monetary policy
variable.
9
5 Estimation
5-1. Estimation Method We employ the TVP-VAR model in a similar manner to Primiceri (2005), Nakajima (2011a)
and Nakajima et al. (2011). Estimations were done by MATLAB, created by Nakajima (2011a)
and arranged by the authors. The TVP-VAR system with lag 𝑠 is given as
𝐴𝑡𝑦𝑡=𝐶1𝑡𝑦𝑡−1 + 𝐶2𝑡𝑦𝑡−2 +⋯+ 𝐶𝑠𝑡𝑦𝑡−𝑠 + 𝜖𝑡 (5)
𝜖𝑡~𝑁(0,Φt), (𝑡 = 𝑠 + 1,…𝑛).
where 𝐶𝑖𝑡 and 𝐴𝑡 are matrixes of (𝑘 × 𝑘) time-varying coefficients (𝑖 = 1,… 𝑠), with
𝑦𝑡=(𝑦1𝑡, … , 𝑦𝑘𝑡)′ and 𝜖𝑡 is a vector of the fundamental structural shocks (𝑘 × 1). Also, Φt is
a variance-covariance matrix (𝑘 × 𝑘)4.
Translating (5), we get reduced form (6) as follows:
𝑦𝑡=𝐵1𝑡𝑦𝑡−1 + 𝐵2𝑡𝑦𝑡−2 +⋯+ 𝐵𝑠𝑡𝑦𝑡−𝑠 + 𝑢𝑡 (6)
𝑢𝑡~𝑁(0, At−1ΦtAt
−1′), (𝑡 = 𝑠 + 1,…𝑛).
where 𝐵𝑖𝑡 = At−1𝐶𝑖𝑡 and 𝐵𝑖𝑡is a matrix of (𝑘 × 𝑘) time-varying coefficient and (𝑖 = 1,… 𝑠),
𝑢𝑡 is an error term vector of (𝑘 × 1).
Then, the variance of 𝑢𝑡 can be reformed with a Cholesky decomposition to impose
recursive restriction:
At−1ΦtAt
−1 = At−1𝛴𝑡𝛴𝑡′At
−1′ (7)
Therefore, the error term 𝑢𝑡 follows the 𝑘 variable normal distribution with an average of
0 and the time-varying covariance matrix Φt. Reorganizing further, we get
𝑦𝑡=𝑋𝑡𝛽𝑡+At−1𝛴𝑡𝑒𝑡 (8)
𝑒𝑡~𝑁(0, I𝑘)13
, (𝑡 = 𝑠 + 1,…𝑛)
where, 𝛽𝑡 = vec〔𝐵1t′, …𝐵st
′〕 and 𝑋𝑡 = 𝐼𝑘⊗(𝑦t−1
′, 𝑦t−2′…𝑦t−s
′) are scaled structural
shocks. At is a lower triangular matrix in which the diagonal elements are equal to one
and 𝛴𝑡 is the diagonal matrix:
𝐴𝑡 = (
1 0 ⋯ 0𝑎21𝑡 1 ⋱ ⋮⋮ ⋱ ⋱ 0𝑎𝑘1𝑡 ⋯ 𝑎𝑘,𝑘−1,𝑡 1
) (9)
13
𝐼 is an identity matrix.
10
𝛴𝑡 = (𝜎1𝑡 ⋯ 0⋮ ⋱ ⋮0 ⋯ 𝜎𝑘𝑡
) (10)
Additionally, 𝑎𝑖𝑗𝑡 is the simultaneous relations of the structural shock parameter that the
structural shock of the variable 𝑗 affects variable 𝑖. Hence, we define the lower triangular
elements of 𝐴𝑡 as 𝑎𝑡= (𝑎21𝑡, 𝑎31𝑡 , 𝑎32𝑡,…,𝑎𝑘𝑘−1𝑡)′.
Converting the diagonal components of 𝛴𝑡 into a single row, it can be represented as
ℎ𝑖𝑡 = log𝜎𝑖𝑡2 . Then, we define ℎ𝑡 =(ℎ1𝑡,…, ℎ𝑘𝑡)′. Here, 𝜎𝑖𝑡
2 is the time-varying variance of the
structural shock of the variable i.
In addition, we assume that the time-varying parameters (𝛽𝑡,𝑎𝑡 , ℎ𝑡) follow the following
random walk process:
𝛽𝑡+1=𝛽𝑡 + 𝑢𝛽𝑡, 𝑎𝑡+1=𝑎𝑡 + 𝑢𝑎𝑡, ℎ𝑡+1=ℎ𝑡 + 𝑢ℎ𝑡. (11)
Here, the error term vector of each variables is as follows:
(
e𝑡𝑢𝛽𝑡𝑢𝑎𝑡𝑢ℎ𝑡)~𝑁
(
𝑜,(
I𝑘 𝑜 𝑜 𝑜𝑜 𝛴𝛽 𝑜 𝑜
𝑜 𝑜 𝛴𝑎 𝑜𝑜 𝑜 𝑜 𝛴ℎ
)
)
, (12)
where it is assumed that (𝛴𝛽 , 𝛴𝑎 , 𝛴ℎ) are time-varying variance.
For sampling, we set the following normal distribution for the initial state of the
time-varying parameters. This can be said to be setting a sufficiently flat prior distribution:
𝛽0~𝑁(0,10𝐼), 𝑎0~𝑁(0,10𝐼), ℎ0~𝑁(0,10𝐼). (13)
We also set the same prior distributions for all models and let the distribution of structural
shocks be well captured. Hence, we determine these prior distributions:
(𝛴𝛽)𝑖2~𝐼𝐺(60, 10−4𝐼), (𝛴𝑎)𝑘
2~𝐼𝐺(6, 2 × 10−2𝐼), (𝛴ℎ)𝑘2~𝐼𝐺(6, 2 × 10−2𝐼). (14)
where (𝛴𝛽)𝑖2 are the 𝑖-th elements of 𝛴𝛽 . Also, (𝛴𝑎)𝑘
2 and (𝛴ℎ)𝑘2 are the 𝑘-th diagonal
elements of 𝛴𝑎 and 𝛴ℎ.
From these prior distributions, the extent of time-varying parameters’ movements is
determined. Tighter prior is set for 𝛽𝑡 than 𝑎𝑡 and ℎ𝑡 to avoid the implausible behaviours of
the time-varying parameters following the way of Nakajima (2011a) with considering the
datasets. This time, two lags are taken in all models because more than three lags may absorb
11
the shock of variance14
. In this paper, we use �̃�𝑖 = 𝛴𝑡=𝑠+1𝑛 exp (ℎ𝑖𝑡/2) for the average volatility
of the structural shock in the sample period.
5-2. MCMC Method In the following, we will explain the way of utilising the Markov chain Monte Carlo
(MCMC) method for sampling. This study conducts a Bayesian estimation using the MCMC
method based on Nakajima (2011a). First, we define
𝑦={𝑦𝑡}𝑡=1𝑛 , 𝛽={𝛽𝑡}𝑡=𝑠+1
𝑛 , 𝑎={𝑎𝑡}𝑡=𝑠+1𝑛 , ℎ={ℎ𝑡}𝑡=𝑠+1
𝑛 and 𝜔 = (𝛴𝛽 , 𝛴𝑎 , 𝛴ℎ). Then, from the
posterior distribution π(𝛽, 𝑎, ℎ, 𝜔|𝑦), we generate the sample from the posterior probability
density function. The steps of MCMC algorithm are as follows:
1. Set initial values of 𝛽,𝑎, ℎ, 𝜔.
2. Sample 𝛽|𝑎, ℎ, 𝛴𝛽 , y.
3. Sample 𝛴𝛽 |𝛽.
4. Sample 𝑎|𝛽, ℎ, 𝛴𝑎 , y.
5. Sample 𝛴𝑎|𝑎.
6. Sample ℎ|𝛽, 𝑎, 𝛴ℎ , y.
7. Sample 𝛴ℎ|ℎ.
8. Go back to 2.
An initial sample of 30,000 is generated and then it is discarded and another sample of
30,000 generated. Next, the sampling frequency is defined as follows. Here, for example,
𝛽|𝑎, ℎ, 𝛴𝛽 , y represents a conditional distribution of 𝛽 conditioned on(𝑎, ℎ, 𝛴𝛽 , y). For further
details on sampling using the MCMC method, see Koop (2003) and Nakajima (2011a).
In the following, Geweke’s CD statistics and the sample autocorrelations of selected
parameters are reported in the tables and figures. The sample autocorrelations listed in Table 3
are the (5, 5) and (15, 15) components of the time-varying parameter 𝛴𝛽, (1,1) and (3,3)
components of 𝛴𝑎, (1,1) and (3,3) components of 𝛴ℎ . As for CD, n0=1,000 and n1=5,000
following Nakajima (2011a). The sample autocorrelation function of these parameters reported
in Table 3 is sufficiently converged. Based on these facts, we can say that the convergence with
30,000 sampling is sufficient.
[Table 3]
[Figure 3]
[Figure 4]
[Figure 5]
14
Most of the results with three lags were almost the same as with two lags.
12
[Figure 6]
[Figure 7]
13
6 Results The results presented in the figures are estimated and drawn by the MATLAB program
created by Nakajima (2011a) and arranged by the authors. Before interpreting the results, we
will explain the figures of impulse response itself. Following Nakajima and Watanabe (2011),
these impulse responses are drawn with posterior medians of the impulse response (solid line,
green) and significant influences (dotted line) of 25% and 75%. For example, 𝜀𝑚 ↑→ 𝑙 looks at
the response of loans when the BOJ’s CAB received a shock. The horizontal axis represents the
March 2000 to March 2016 period and the vertical axis shows the size of the response. Focusing
on the horizontal axis, the sample period is divided into three periods: the first break is March
2006 and the second break is March 2013. This is determined while considering the policy
terms and policy objectives. The figures are arranged to see the impulse response after 6 months
to see the short-term and 2 years to see the long-term. From these figures, it is possible to check
the impact of monetary policy shocks by time and size of firms.
6-1. Result interpretation of Model 1 Looking at the results of Model 1(L,Y,M) (Figs.8–10), effects of monetary easing shock on
loans (M→L) differ by the firm size. Specifically, large firms have positive effect after 2010 but
no effect in other periods. Middle-sized firms have no significant effect and small firms have a
shifted effect from short-term negative to long-term positive. The reason why short-term and
long-term effects of monetary policy on loans (M→L) are different in small firms may well be
due to portfolio rebalance effects. Another noteworthy point is that the effect from loan shock to
IIP (L→Y) is always negative in large and middle-sized firms. This can be thought that these
firms are issuing corporate bonds, but do not borrow from banks.
[Figure 8]
[Figure 9]
[Figure 10]
6-2. Result interpretation of Model 2 Next, to take into account the possibility that loan stagnation in Model 1 was caused by the
supply side of loans, we incorporate BCC to capture fluctuation in the lending standard. In
Model 2 (BCC,L,Y,M), we will look at the effects of M→BCC, BCC→L and L→Y.
Looking at the effect from easing shock to BCC(M→BCC) and the effect of BCC shock on
loan(BCC→L), we can say that even if the banks’ lending standard (BCC) is eased, loans do not
increase in large and middle-sized firms. The lending standard affects the loan volume only for
small firms. This is probably because BCC can rise due to corporate factors when the corporate
performance is good, which leads to a decrease in loan demand. A more detailed study of this
14
hypothesis will be considered in models that include loan demand (CCC), bank factors (BIC)
and corporate factors (BOC)15
. Moreover, taking the effect of L→Y into consideration, the
channel to output through BCC increased output through loan increase (BCC→L→Y), but only
in the short term of small firms. To sum up, the effect of monetary policy through the credit
channel is limited; however, large and middle-sized firms have a negative effect on output from
increasing loans. This is probably because the portfolio rebalancing effect works strongly in
large and middle-sized firms. Thus, although easing lending standards has a positive effect on
output, it is not thought to have an effect through loans.
[Figure 11]
[Figure 12]
[Figure 13]
6-3. Result interpretation of Model 3 In previous models, it is also possible to consider that the easing of shocks’ low effects on
loans is caused by low loan demand in the first place. Therefore, we will consider Model
3(CCC,L,Y,M). Here, we are going to look at M→CCC and CCC→L.
Focussing on M→CCC, the effect from easing shock on loan demand (M→CCC) is positive
for almost all firm sizes and policy periods. This means that monetary easing certainly affects
corporate activities or corporate finance positively. The effect of loan demand shock on loans
(CCC→L) is positive or insignificant in large and small firms. On the other hand, in
middle-sized firms, the effect is negative in some periods. This seems strange that an increase in
loan demand leads to a decrease in loans. This strange result may be because the financial
institutions might have concerns about unstable corporate management and do not increase
loans even CCC rises. Considering these results, it can be said that increase in the loan demand
(CCC) has a positive effect on loan(L) even if it is different in degree.
[Figure 14]
[Figure 15]
[Figure 16]
6-4. Result interpretation of Model 4 Considering the results so far, we can see that the demand for loans is rising in response to
monetary easing. This means that the hypothesis we set up in Model 2, ‘even if BCC rises, loans
15
As another interpretation, there is a possibility that BCC has relaxed the decrease of loan (L).
15
are sluggish due to decrease in loan demand’, turns out to be wrong. Therefore, the reasons for
sluggish loan volume are considered with classifying BCC as bank factor (BIC) and corporate
factor (BOC).
First, Model 4 (BIC,L,Y,M) can clearly consider the possibility that lending has been
sluggish due to bank-side problems such as bad loans and low capital adequacy ratio. Here, we
are going to look at M→BIC, BIC→L, BIC→Y and L→Y.
Looking at M→BIC and BIC→L, it is possible to say that banking assets might have
affected loans to small firms. The reason why the effect of BIC on L(BIC→L) changes is
thought to be because bad loan disposal is ongoing in the first policy term until March 2006. In
addition, considering IIP while looking at BIC→Y, it was confirmed that the channel of bank
assets, or BIC is functioning in all firm sizes16
. Also, since L→Y has a negative effect in large
and middle-sized firms, we can consider the possibility that IIP has increased through the rise in
the stock price due to the portfolio rebalancing effect. To sum up, the results of this model tells
us that the channel of loan increase through bank assets did not work effectively for large and
middle-sized firms. On the other hand, in small firms, this channel seems have worked only for
a certain period. However, since the effect of monetary easing shock on BIC (M→BIC) for
small firms is weak, it is considered that the effect of BIC→L→Y is not large. Therefore, the
bank factor (BIC) explains some of the reasons why lending is sluggish in large and
middle-sized firms even though the lending standard (BCC) is eased.
[Figure 17]
[Figure 18]
[Figure 19]
6-5. Result interpretation of Model 5 In Model 5 (BOC,L,Y,M), we will focus on the effects of M→BOC, BOC→L and L→Y.
From the results of M→BOC, monetary easing policy improves the business outlook for
large and small firms. Based on this, it is conceivable that some small firms are the subsidiaries
of large firms; on the other hand, middle-sized firms do not have sufficient sales channels in
overseas nor their managements are becoming unstable due to the exchange rate fluctuations.
Indeed, for middle-sized firms, easing shock on IIP (M→Y) effects are negative17
. Next, looking
at BOC→L, this effect is basically negative for large and middle-sized firms, whereas small
firms have a positive effect in second and third policy terms. This is thought to be because large
16
Using bank stock prices, Harada and Masujiima (2008) analyse banks’ balance sheet channels between
2001 and 2006. They conclude that this channel worked in their sample period based on the results that
easing shock has a positive effect on the bank stock price and the bank stock price has a positive effect on
the production. 17
This result is not posted in this paper.
16
and middle-sized firms can procure their own funds with stocks and corporate bonds when the
business outlook improves. In this regard, the channel of BOC on loans is effective only for
small firms. At the same time, however, this channel seems to have very weak effect for small
firms considering the effect of monetary easing shock on BOC (M→BOC). In addition,
focussing on L→Y, this has a negative effect for large and middle-sized firms. It seems that
there is a high possibility that the portfolio rebalancing effect was working. Considering these
results, it is conceivable that corporate factors effectively explain the reasons why loans are
sluggish in large and middle-sized firms even if the lending standard is eased.
[Figure 20]
[Figure 21]
[Figure 22]
17
7 Conclusion Below, we review and summarise five models in this sample period. At the end, we then will
consider the importance of each transmission channel.
First, we focus on large firms. As the results of this paper show, large firms might have
increased IIP through the effect on stock prices caused by portfolio rebalancing. As another
hypothesis, we can consider that even if loans to large firms increase, large firms can decrease
their output to prepare for corporate finance such as exchange rate fluctuations, or it is also
possible to think that loans are devoted to overseas investment and, therefore, are not much
reflected in the Japanese IIP. Regarding this possibility, we should consider how firms finance
overseas investment, but we set this as a future task. At the same time, we also have to consider
the possibility that loans from domestic banks to the foreign branches cause a decrease in the
accuracy of estimation in this paper.
Second, it seems that middle-sized firms are unlikely to have a favourable effect by
monetary easing policy; rather, they are likely to be influenced negatively due to the instability
of management18
. The reason for this might be because the development of overseas sales is not
substantial and preparation for exchange rate fluctuations is not sufficient. Indeed, in most of the
results, middle-sized firms have a negative shock from monetary easing on IIP.
Regarding small firms, the effect on IIP through loans tends to be positive compared with
other firm sizes. We observed in the long term that easing shock has a positive effect on loan to
small firms. However, loans will not grow at the beginning maybe because the management of
small firms is not stable. Then, as the performance of small firms becomes stable, loans increase
to invest or produce more goods.
To summarise the above points, it can be confirmed that the effect of monetary easing policy
on loans and the effect of loan increase on output are different depending on the firm size. This
can be because the ways of financing overseas investment and the components of assets—the
composition of the balance sheet and the degree of dependence on financial institutions—are
different depending on the firm size. The detailed reason as to why the effect of monetary easing
differed by the firm size should be considered by looking at more specific portfolio rebalancing
effects, which we set as a future task19
. One more important thing to be mentioned is that the
effect on IIP through loans tends not to work positively except for small firms.
In conclusion, we will summarize the importance of each channel.
First, regarding BCC, this channel does not work for large and middle-sized firms. On the other
hand, for all firm sizes, CCC had been effective for some periods. Therefore, the reason why
loans are sluggish is not explained by demand for loans, but by the bank factors and the
18
Regarding the case of Tokyo, the survey about the effect of exchange rate on middle- or small-sized
firms’ managements are uploaded on the website of The Tokyo Chamber of Commerce and Industry. 19
Saito and Hogen (2014) investigated the details of portfolio rebalancing and mentioned that the
Japanese domestic banks have reacted to the Quantitative and Qualitative Monetary Easing (QQE) policy
by rebalancing their portfolios; increasing loans and investment in equities.
18
corporate factors within the lending standard. In detail, focussing on the bank factors (BIC), we
can say that even if BIC rises, it will not lead to a rise in loans for large and middle-sized firms.
For small firms, however, loans will increase in the second and third policy terms even this
effect is thought to be very weak. In this respect, the effect of bank factors improving loans
seems to be small. Next, looking at corporate factors (BOC), only for small firms, loans increase
as BOC improves in response to easing shock. This means that the channel through BOC is
effective only in small firms. However, as well as BIC, this channel seems to have a very weak
effect. Therefore, the reason why the credit channel does not work is explained by BIC and
BOC and improvement of these factors does not increase loans and even if it grows, the effect is
limited. In summary, it is possible to say that (1) Japanese monetary easing policy had little
influence on Japanese output through loans; (2) fluctuation in loan volume is largely explained
by the lending standard more than a change in loan demand; (3) broadly speaking, bank factors
and corporate factors within the lending standard do not have a positive effect on loans even if
they react to monetary easing shock and (4) the channel can be working effectively only for
small firms even if the effect is considered to be small.
A more detailed approach on loans and portfolio rebalancing is needed to analyse this with
better precision, but these implications are important for central banks when considering the
effects of monetary easing policy on loans from the perspective of banks and figuring this
channel out from macro-perspective.
Acknowledgements
We are grateful for helpful comments from Toshiki Jinushi, Tomomi Miyazaki (Kobe
University) and Hiroyuki Ijiri (Okayama Shoka University). We are also grateful for valuable
comments and seminar participants at the Kochi seminar. Any remaining errors are our
responsibility.
19
References
Bassett, W. F., Chosak, M. B., Driscoll, J. C., & Zakrajšek, E. (2014). “Changes in bank lending
standards and the macroeconomy”. Journal of Monetary Economics, 62, 23-40.
Bernanke, B.S, Blinder, A.S, (1988). ”Credit, money and aggregate demand”. The American
Economic Review, 78, 435–439.
Bernanke, B. S., & Blinder, A. S. (1992). ”The federal funds rate and the channels of monetary
transmission”. The American Economic Review, 901-921.
Bernanke, B. S. (1993). ”Credit in the Macroeconomy”. Quarterly Review-Federal Reserve
Bank of New York, 18, 50-50.
Bernanke, B. & Gertler, M. (1995). “Inside the black box: the credit channel of monetary policy
transmission”, Journal of Economic Perspectives, vol.9, pp. 27-48.
Bernanke, B. S., & Reinhart, V. R. (2004). “Conducting monetary policy at very low short-term
interest rates”. The American Economic Review, 94(2), 85-90.
Christiano, Lawrence J., Martin Eichenbaum, and Charles L. Evans. (1999). "Monetary policy
shocks: What have we learned and to what end?". Handbook of macroeconomics 1: 65-148.
Ciccarelli, M., Maddaloni, A., & Peydro, J. L. (2015). “Trusting the bankers: A new look at the
credit channel of monetary policy”. Review of Economic Dynamics, 18(4), 979-1002.
Del Giovane, P., Eramo, G., & Nobili, A. (2011). “Disentangling demand and supply in credit
developments: a survey-based analysis for Italy”. Journal of Banking & finance, 35(10),
2719-2732.
Honda, Y., Kuroki, Y., & Tachibana, M. (2007). “An injection of base money at zero interest
rates: empirical evidence from the Japanese experience 2001-2006”. Graduate School of
Economics and Osaka School of International Public Policy. Discussion Paper 07-08.
Ijiri, H. (2016). “Re-evaluating Japan’s Quantitative Easing Policy (2001–2006): An Application
of the TVP-VAR Model”. Japanese Journal of Monetary and Financial Economics Vol, 4(1),
1-17.
Iwata, S., & Wu, S. (2006). “Estimating monetary policy effects when interest rates are close to
zero”. Journal of Monetary Economics, 53(7), 1395-1408.
Kano, S. (2006) Macroeconomic Analyses and Survey Data (in Japanese), (Iwanami Shoten).
Kashyap, A. K., & Stein, J. C. (1994). “Monetary policy and bank lending”. In Monetary policy
(pp. 221-261). The University of Chicago Press.
Kimura, T., & Nakajima, J. (2016). Identifying conventional and unconventional monetary
policy shocks: a latent threshold approach. The BE Journal of Macroeconomics, 16(1),
277-300.
Koop, G. (2003). Introduction to Bayesian Econometrics.
Lown, C., & Morgan, D. P. (2006). “The credit cycle and the business cycle: new findings using
the loan officer opinion survey”. Journal of Money, Credit and Banking, 1575-1597.
Lown, C. S., Morgan, D. P., & Rohatgi, S. (2000). “Listening to loan officers: The impact of
commercial credit standards on lending and output”. Economic Policy Review, 6(2).
20
Maddaloni, A., & Peydró, J. L. (2011). “Bank risk-taking, securitization, supervision, and low
interest rates: Evidence from the Euro-area and the US lending standards”. Review of
Financial Studies, 24(6), 2121-2165.
Harada, Y. & Masujima, M. (2008). “Effectiveness and Transmission Mechanism of the
Quantitative Monetary Easing Policy” (in Japanese), ESRI Discussion Paper Series, No.204
Nakajima, J. (2011a). “Time-varying parameter VAR model with stochastic volatility: An
overview of methodology and empirical applications”. Monetary and Economic Studies,
29,107–142.
Nakajima, J. (2011b). “Monetary policy transmission under zero interest rates: An extended
time-varying parameter vector autoregression approach”. The BE Journal of Macroeconomics,
11(1).
Nakajima, J., Kasuya, M., & Watanabe, T. (2011). “Bayesian analysis of time-varying parameter
vector autoregressive model for the Japanese economy and monetary policy”. Journal of the
Japanese and International Economies, 25(3), 225-245.
Nakajima, J., & Watanabe, T. (2011). “Bayesian analysis of Time-Varying Parameter Vector
Autoregressive model with the ordering of variables for the Japanese economy and monetary
policy”. Global COE Hi-Stat Discussion Paper Series, 196.
Nakajima, J., & West, M. (2013). “Bayesian analysis of latent threshold dynamic models”.
Journal of Business & Economic Statistics, 31(2), 151-164.
Newey, W. K., & West, K. D. (1987). “Hypothesis testing with efficient method of moments
estimation”. International Economic Review, 777-787.
Ogawa, K. (2003). “Financial distress and employment: the Japanese case in the 90s”, National
Bureau of Economic Research, No.w9646.
Oliner, S. D., & Rudebusch, G. D. (1996). “Is there a broad credit channel for monetary
policy?”. Economic Review-Federal Reserve Bank of San Francisco, (1), 3.
Primiceri, G. E. (2005). “Time varying structural vector autoregressions and monetary policy”.
The Review of Economic Studies, 72(3), 821-852.
Saito, M., & Hogen, Y. (2014). “Portfolio Rebalancing Following the Bank of Japan's
Government Bond Purchases: Empirical Analysis Using Data on Bank Loans and Investment
Flows”. BOJ Reports & Research Papers, Bank of Japan.
Uchida M. (2013). “An Effect and Limit of an Unconventional Monetary Policy: Overcome of
Deflation and a Monetary Policy.” (in Japanese), Annual report of the Institute for Economic
Studies, 26. ,129-160.
Williams, J. C. (2012). “The Federal Reserve’s Unconventional Policies”. FRBSF Economic
Letter, 34, 1-9.
21
Appendix: Tables and Figures.
Table 1: Transmission Channel Variables
Variable Purpose
CCC or demand
(cost of capital channel)
To express the loan demand from firms that banks recognise.
BCC
(broad credit channel)
To express changes of lending standards to firms that banks recognise.
BIC
(bank incentive channel)
To express the volume of financial institutions’ sentiment; i.e. how much
banks’ assets (or balance sheets) affect the lending standard.
BOC
(bank observe channel)
This variable expresses the volume of financial institutions’ sentiment; i.e. how
much is the business outlook for firms affected by the lending standard.
Table 2: Summary of Data and Sources
Data Description and Source
IIP(Y) Industrial production index. These data are obtained from the website of the Ministry of
Economy, Trade and Industry (seasonally adjusted series, 2010 average = 100).
BOJ current account
balances(M)
These data are obtained from the BOJ’s website:
Monetary base/current account balances/average amounts outstanding.
The data are seasonally adjusted by E-views.
Loans and bills
discounted by sector
(by scale of
enterprises) (L)
Loan data are obtained from the BOJ’s website except for SME_CAL.
Detailed information is on the BOJ’s website:
https://www.boj.or.jp/en/statistics/outline/exp/exyo.htm/.
The following notes are taken from the BOJ’s website: ‘These data include (1) Domestically
licensed banks (banking accounts, trust accounts and overseas office accounts), (2) Shinkin
Banks (banking accounts); (3) Other financial institutions (banking accounts), e.g. the
Norinchukin Bank, the Shoko Chukin Bank, Development Bank of Japan, Japan Finance
Corporation (Micro Business and Individual Unit, Small and Medium Enterprise Unit and
Agriculture, Forestry, Fisheries and Food Business Unit), Japan Bank for International
Cooperation and the Okinawa Development Finance Corporation’.
L:BIG and L:MID Large or middle-sized enterprises/corporations including financial corporations, outstanding
bank accounts, trust accounts and overseas office accounts and domestically licensed banks.
These data are seasonally adjusted by E-views.
L:SME_CAL Loans for small enterprises (SME_CAL) are calculated by the authors as (SME_CAL) =
(ALL) − (BIG) − (MID). This is seasonally adjusted by E-views after this calculation.
Bank loan amounts of small firms existed at the time of April 2016, but there was a
subsequent change in the dataset of the BOJ. Calculations based on the above formula
showed that there was almost no difference between these values.
Channel variables Data from the ‘Senior Loan Officer Opinion Survey on Bank Lending Practices at Large
22
created from
SLOS(C)
Japanese Banks’ survey is obtained from the BOJ website. The authors then calculated and
seasonally adjusted by E-views. An explanation of SLOS is here:
https://www.boj.or.jp/en/statistics/outline/notice_2000/ntloos01.htm/.
According to the BOJ citation, the aggregated loan amount of the surveyed 50 banks
accounts for about 75 percent of the total amount outstanding of loans held by all
domestically licensed banks and shinkin banks (the average during fiscal 2015). The
questionnaire target is reconsidered every 3 years based on the total amount of loans and was
changed in April 2003, 2006, 2009, 2012 and 2015.
Table 3: Geweke’s CD statistics (p-value)
Parameter (𝛴𝛽)5 (𝛴𝛽)15 (𝛴𝑎)1 (𝛴𝑎)3 (𝛴ℎ)1 (𝛴ℎ)3
Model 1 Large firms 0.905 0.382 0.665 0.062 0.855 0.117
Model 1 Middle firms 0.543 0.397 0.190 0.919 0.112 0.443
Model 1 Small firms 0.950 0.103 0.398 0.219 0.218 0.475
Model 2 Large firms 0.601 0.372 0.999 0.643 0.276 0.980
Model 2 Middle firms 0.549 0.941 0.243 0.159 0.775 0.717
Model 2 Small firms 0.948 0.096 0.493 0.718 0.254 0.456
Model 3 Large firms 0.983 0.191 0.557 0.589 0.479 0.658
Model 3 Middle firms 0.874 0.214 0.208 0.112 0.501 0.733
Model 3 Small firms 0.739 0.610 0.073 0.393 0.782 0.988
Model 4 Large firms 0.755 0.381 0.060 0.186 0.516 0.718
Model 4 Middle firms 0.820 0.751 0.717 0.840 0.649 0.997
Model 4 Small firms 0.391 0.045 0.547 0.060 0.294 0.386
Model 5 Large firms 0.409 0.403 0.925 0.160 0.060 0.174
Model 5 Middle firms 0.117 0.136 0.627 0.200 0.212 0.481
Model 5 Small firms 0.521 0.959 0.753 0.695 0.093 0.697
23
Figure1: Expected framework of transmission channels.
Note: Figures 1 is created by the authors based on Uchida (2013) and Ciccarelli et al. (2015).
Figure2: The movements of BCC and Lending Attitude.
Note: The scale of LA is on the right-side and BCC on left-side. Lending Attitude is observed
and calculated by BOJ (see the BOJ’s website for detailed information).
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
200
0/3/
31200
0/9/
29200
1/3/
30200
1/9/
28200
2/3/
29200
2/9/
30200
3/3/
31200
3/9/
30200
4/3/
31200
4/9/
30200
5/3/
31200
5/9/
30200
6/3/
31200
6/9/
29200
7/3/
30200
7/9/
28200
8/3/
31200
8/9/
30200
9/3/
31200
9/9/
30201
0/3/
31201
0/9/
30201
1/3/
31201
1/9/
30201
2/3/
30201
2/9/
28201
3/3/
29201
3/9/
30201
4/3/
31201
4/9/
30201
5/3/
31201
5/9/
30201
6/3/
31201
6/9/
30
BCC and Lending Attitude (small firms)
BCC_SMA LA_SMA
24
Figure3: Sample autocorrelations of Model 1 for selected parameters.
Figure4: Sample autocorrelations of Model 2 for selected parameters.
Figure5: Sample autocorrelations of Model 3 for selected parameters.
25
Figure6: Sample autocorrelations of Model 4 for selected parameters.
Figure7: Sample autocorrelations of Model 5 for selected parameters.
Figure8: Model 1 (large firms)
26
Figure9: Model 1 (middle-sized firms)
Figure10: Model 1 (small firms)
Figure11: Model 2 (large firms). (For simplicity, BCC is indicated as ‘bc’ in Figs.11–13.)
27
Figure12: Model 2 (middle-sized firms)
Figure13: Model 2 (small firms)
Figure14: Model 3 (large firms). (For simplicity, CCC is indicated as ‘cc’ in Figs.14–16.)
28
Figure15: Model 3 (middle-sized firms)
Figure16: Model 3 (small firms)
Figure17: Model 4 (large firms) (For simplicity, BIC is indicated as ‘bi’ in Figs.17–19.)
29
Figure18: Model 4 (middle-sized firms)
Figure19: Model 4 (small firms)
Figure20: Model 5 (large firms). (For simplicity, BOC is indicated as ‘bo’ in Figs. 20–22.)
30
Figure21: Model 5 (middle-sized firms)
Figure22: Model 5 (small firms)
Recommended